Pith. sign in

REVIEW 3 major objections 6 minor 88 references

Towards Unified Neurosymbolic Reasoning on Knowledge Graphs

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read TUNSR is a unified neurosymbolic framework that performs both propositional and first-order logic reasoning for knowledge-graph link prediction, reporting state-of-the-art or competitive results across 19 datasets in four reasoning…

desk verdict A serious empirical unification of four KG reasoning scenarios, with the neurosymbolic framing stronger than the formal support; worth refereeing, but the authors need to fix Eq. (13) and tone down what 'FOL reasoning' means. read the letter →

arxiv 2507.03697 v1 pith:KCFNSUCN submitted 2025-07-04 cs.AI

Qika Lin , Fangzhi Xu , Hao Lu , Kai He , Rui Mao , Jun Liu , Erik Cambria , Mengling Feng This is my paper
classification cs.AI
keywords knowledgegraphreasoningneurosymboliclinkpredictionfirst-orderlogicrulespropositionaltemporalgraphsruleinductioninterpretability
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

TUNSR is a single neurosymbolic architecture for knowledge-graph link prediction that claims to combine propositional reasoning (entity-dependent multi-hop path scoring) with first-order logic reasoning (entity-independent rule application) in one differentiable model. The paper argues that both reasoning styles can be expressed over a uniformly constructed "reasoning graph" grown from the query entity, so that one network can serve transductive, inductive, interpolation, and extrapolation settings without architectural changes. It further claims that first-order rule confidence can be computed by merging possible relations step by step rather than by enumerating every rule grounding, and that the resulting attention weights can be read back into explicit, human-readable logic rules. The reported results on 19 datasets across the four scenarios, together with case studies and ablations, are offered as evidence that the unified mechanism is accurate as well as interpretable.

What carries the argument

The load-bearing object is the reasoning graph, a tree rooted at the query entity (or query entity-time pair for temporal graphs) that expands by searching posterior neighbors and carries a self-loop relation so a node can persist across iterations. Over this graph, a forward logic message-passing mechanism runs two parallel streams: propositional message passing computes entity-dependent node embeddings and attention scores via sigmoid and softmax operations, while FOL message passing uses a GRU to fuse the current relation with the preceding node's FOL embedding and computes entity-independent attention. The central identity is Eq. (13), the asserted approximation that the score obtained by aggregating confidences of all grounded FOL rules equals the score obtained by step-wise, attention-based merging of possible relations; this is what lets rule confidence be accumulated online. The FARI algorithm closes the loop by multiplying FOL attentions along graph paths, adding relations to rule bodies, and deleting self-loop atoms to enumerate all possible Horn rules up to length L.

What would settle it

On a small knowledge graph, train TUNSR, then for a set of queries enumerate every grounded FOL rule that the FARI algorithm can induce, compute the exact confidence sum from part (a) of Eq. (13), and compare it with the step-wise score from part (b); a large or systematic gap would show that the induced rules are not doing the scoring work the paper attributes to them. A second check is to ablate the FOL stream and ask whether the highest-confidence induced rules still predict the same answers, which would separate genuine rule usage from post hoc explanation.

Watch

Extended reading notes

Core claim

On its own terms, the central discovery is that a single network can carry out both propositional and first-order logic reasoning for knowledge-graph link prediction if the two streams are defined over one expanding reasoning graph. The propositional stream updates node embeddings and attentions from entity-dependent messages, while the FOL stream passes GRU-composed relation embeddings from node to node and accumulates entity-independent attention that acts as rule confidence. The paper's key claimed equivalence is that summing the confidences of all grounded FOL rules (part (a) of Eq. (13)) can be approximated by a step-wise process that merges possible relations at each hop (part (b)), making rule learning and grounding differentiable. The FARI algorithm then induces explicit Horn rules — connected and closed rules for static graphs, plus temporal interpolation and extrapolation variants — from the learned attentions, so predictions come with a readable chain of rules. The experiments across 19 datasets and four reasoning scenarios are presented as evidence for both the accuracy and the interpretability of this unified design.

Load-bearing premise

The load-bearing premise is that the step-wise, attention-based merging of possible relations in part (b) of Eq. (13) faithfully approximates summing the confidences of every grounded first-order logic rule in part (a); the paper marks this as an approximation without a proof, and if the equivalence fails, the induced rules and attention scores would not be the true drivers of the model's predictions.

Editorial extensions

If this is right

  • A single TUNSR model answers transductive, inductive, interpolation, and extrapolation queries without changing the architecture; only the node structure of the reasoning graph changes.
  • The attention weights used for scoring double as rule confidences, so every prediction can be accompanied by the chained FOL rules that produced it.
  • Because FOL reasoning is entity-independent, the induced rules transfer to unseen entities and unseen graphs that share the relation set, which is what makes inductive reasoning possible.
  • Combining propositional and FOL streams outperforms either alone across the reported settings, since propositional paths capture entity-specific evidence while FOL rules provide generalization.
  • Rule length behaves as a bias-variance knob: performance improves up to roughly 6 hops on static and 4 hops on temporal graphs, then stabilizes or degrades as longer rules add noise.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the Eq. (13) approximation holds beyond these benchmarks, the same "merge relations instead of enumerating groundings" recipe could be applied to any structured scoring task that is a sum over grounded rules, such as multi-hop question answering or program synthesis over graphs.
  • The separate propositional and FOL attention maps enable a diagnostic not reported in the paper: on a failure case, one can check whether the error came from missing path evidence (propositional attention near the correct entity but low FOL confidence) or from over-general rules (high FOL confidence but wrong entity), which would turn interpretability into a debugging tool.
  • A human study could test the interpretability claim directly: if humans asked to simulate the model using only the induced rules cannot predict its answers, the rules would be post hoc, an outcome the paper's case studies do not rule out.
  • The learned mixing weight λ in Eq. (14) may itself be a useful dataset statistic; reporting it per scenario could reveal a measurable link between graph density and whether propositional or FOL reasoning dominates.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes TUNSR, a graph-based neurosymbolic framework for knowledge graph link prediction that targets transductive, inductive, interpolation, and extrapolation reasoning. It constructs a query-specific expanding reasoning graph, runs forward message passing that tracks both propositional (entity-dependent) embeddings/attentions and first-order-logic (entity-independent) embeddings/attentions, and introduces a Forward Attentive Rule Induction (FARI) algorithm that extracts Horn rules from the learned FOL attentions. The paper reports state-of-the-art or competitive results on 19 datasets, with ablations, hyperparameter studies, and case studies. The central conceptual claim is that the attention-based computation is equivalent to differentiable merging of grounded FOL rules, which is formalized by the approximation in Eq. (13).

Significance. If the theoretical link were established, TUNSR would be a significant contribution: one architecture spanning four reasoning scenarios with rule-level interpretability and consistently strong empirical performance. The experimental effort is substantial and includes many baselines, 19 datasets, ablation and hyperparameter analyses, and concrete rule examples, and the authors are transparent about the inductive-scenario limitation. The significance is contingent, however, because the equivalence in Eq. (13) and the fidelity of FARI's extracted rules to the model's actual scoring function are not demonstrated; without them, the neurosymbolic interpretation is a post-hoc label rather than a derived property of the model.

major comments (3)
  1. [Section 3.3, Eq. (13)] The approximation \(S_o = \sum_{\gamma\in\Gamma}\beta_\gamma \approx \prod_{l=1}^L \sum_{n_j\in O_l} \bar f_l(\mathrm{GRU}(g_{\bar r}, o_l^{n_j}))\) is load-bearing for the claim that TUNSR performs FOL reasoning, but it is asserted without derivation. On the left, the sum is over grounded rules whose bodies are connected paths ending at the target entity \(o\); on the right, the product of independent sums over all nodes in each layer includes combinations of nodes that are not connected by edges in the reasoning graph, and the expression does not restrict the final-layer node to be \(o\). This is not a harmless shorthand, and the paper needs either a proof with the appropriate connectivity constraints or a careful statement of the additional assumptions under which the approximation holds.
  2. [Section 3.3, Algorithm 1, Eq. (15)] FARI multiplies FOL attention values along paths to compute rule confidences, but the final prediction score in Eq. (15) combines propositional and FOL terms through the learned weight \(\lambda\) and a linear layer \(W_5\). The manuscript never shows that the sum of FARI's extracted rule confidences equals or approximates the score used for ranking, so the induced rules in Tables 7-8 are not demonstrably the model's reasoning mechanism. A fidelity experiment, such as ranking candidates using only FARI-derived rule scores and comparing with TUNSR's full rankings, would directly address this gap.
  3. [Section 4.3] The text states 'As inductive reasoning is entity-independent, we only conduct experiments using FOL reasoning for it.' This admitted limitation conflicts with the paper's headline claim of a unified propositional-and-FOL model across all four reasoning scenarios. The authors should clarify whether the inductive results in Tables 3-4 use the full TUNSR or the FOL-only configuration, and should ideally include the propositional and unified variants in the inductive ablation as well.
minor comments (6)
  1. [Eq. (13)] The symbol \(o_l^{n_j}\) in part (b) is not defined; it should presumably be the FOL embedding \(y_{n_j}^l\) of the node.
  2. [Section 3.3] The text refers to 'DURM [32]' but the cited work is DRUM; please correct the typo.
  3. [Table 5] The caption says the table includes ICEWS14, ICEWS0515, and ICEWS18, but only ICEWS14 and ICEWS0515 columns appear; please reconcile the caption with the table.
  4. [Section 4.2] The phrase '14,17% Hits@10 improvement' contains a comma where a decimal point is intended; please correct.
  5. [Algorithm 1] The algorithm's notation \(e^2_{i,l-1}\) and \(e^2_{n^{l-1}_i,\bar r,n^l_j}\) is not connected to the \(\beta\) attention variables defined in Eqs. (10)-(12), making the rule-confidence update hard to follow.
  6. [Abstract and Section 1] The model name is rendered inconsistently as 'TUNSR', 'T UNSR ', and 'Tunsr'; please standardize the spelling.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical claims rest on independent benchmarks, and the FARI rule induction is a post-hoc attention decomposition rather than a fitted input to prediction.

full rationale

TUNSR's scores are computed from learned entity/relation/time embeddings and attention weights via Eqs. (4)-(15), optimized by the multi-class log-loss in Eq. (16). The induced FOL rules in Algorithm 1 are products of the same FOL attentions used in forward scoring, but the paper does not feed these rules back into training or use them to compute the training loss; they are extracted after training as explanations. Thus the prediction does not reduce to the induced rules by construction. The only self-citation is the provenance statement that TUNSR extends the authors' TECHS model [30], which is not used to justify the correctness of the new method. The approximation in Eq. (13), where the sum over grounded FOL rules is identified with a step-wise attention product, is asserted with an unproven '≈' and could be wrong; that is a correctness or evidence gap, not circularity, because the model's actual scoring is defined by the attention product regardless of whether the rule-sum interpretation holds. No parameter is fitted to a target and then renamed as a prediction, and no uniqueness or ansatz is imported from prior self-cited work. The paper's effectiveness claim is supported by external comparisons on 19 datasets, so the derivation chain is self-contained.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The ledger is modest: no physical or conceptual entities are invented. The model introduces a reasoning graph and attention-based rule induction, but these are computational constructs, not new postulates about the world. Key free parameters are hyperparameters (L, M, N) and learned weights (lambda, time encoding parameters). The most significant ad hoc assumption is the factorization approximation in Eq. (13).

free parameters (5)
  • L = Varies per dataset (e.g., 4, 6, 8)
    Number of reasoning iterations / rule length; chosen by hand based on validation performance (Section 4.4).
  • M = Values in {50, 100, 200, 600, 800, 1000}
    Maximum number of posterior neighbor nodes sampled per iteration in extrapolation; tuned for performance versus GPU memory (Section 4.4).
  • N = Values in {40, 60, 80, 100, 120, 140}
    Number of edges retained after attention pruning; tuned in Section 4.4.
  • lambda = Learned during training
    Learnable combination weight between propositional and FOL reasoning in Eq. (14); a free parameter in the fusion.
  • Time encoding parameters w and b = Trainable
    Weights and biases in the generic time encoding used for temporal node embeddings (Section 3.2).
assumptions (4)
  • domain assumption Horn rules are sufficient to model all relevant KG reasoning patterns.
    The paper restricts all FOL rules to Horn rules (Section 2.2), which may limit expressiveness for some queries, but this restriction is standard in the field.
  • domain assumption The reasoning graph expansion, with sampling for extrapolation, preserves the information needed for accurate prediction.
    The graph is pruned and sampled for efficiency (Section 3.4); the paper assumes the sampled paths are sufficient, which is a modeling choice.
  • ad hoc to paper The approximation in Eq. (13), that step-wise relation merging via attention equals summing grounded rule confidences, is valid.
    This equivalence is asserted with a '≈' symbol in Section 3.3 and is not proven; the neurosymbolic interpretation and the FARI algorithm depend on it.
  • domain assumption Time embeddings from the generic time encoding capture temporal semantics for entity-time nodes.
    The paper uses a fixed-form time encoding and assumes it adequately represents temporal facts for interpolation and extrapolation reasoning.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Unified Neurosymbolic Reasoning on Knowledge Graphs." pith.science (2026). https://pith.science/paper/KCFNSUCN

@misc{pith2026250703697,
  author       = {Pith},
  title        = {Pith review of: Towards Unified Neurosymbolic Reasoning on Knowledge Graphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KCFNSUCN}},
  note         = {Machine review of arXiv:2507.03697}
}
read the original abstract

Knowledge Graph (KG) reasoning has received significant attention in the fields of artificial intelligence and knowledge engineering, owing to its ability to autonomously deduce new knowledge and consequently enhance the availability and precision of downstream applications. However, current methods predominantly concentrate on a single form of neural or symbolic reasoning, failing to effectively integrate the inherent strengths of both approaches. Furthermore, the current prevalent methods primarily focus on addressing a single reasoning scenario, presenting limitations in meeting the diverse demands of real-world reasoning tasks. Unifying the neural and symbolic methods, as well as diverse reasoning scenarios in one model is challenging as there is a natural representation gap between symbolic rules and neural networks, and diverse scenarios exhibit distinct knowledge structures and specific reasoning objectives. To address these issues, we propose a unified neurosymbolic reasoning framework, namely Tunsr, for KG reasoning. Tunsr first introduces a consistent structure of reasoning graph that starts from the query entity and constantly expands subsequent nodes by iteratively searching posterior neighbors. Based on it, a forward logic message-passing mechanism is proposed to update both the propositional representations and attentions, as well as first-order logic (FOL) representations and attentions of each node. In this way, Tunsr conducts the transformation of merging multiple rules by merging possible relations at each step. Finally, the FARI algorithm is proposed to induce FOL rules by constantly performing attention calculations over the reasoning graph. Extensive experimental results on 19 datasets of four reasoning scenarios (transductive, inductive, interpolation, and extrapolation) demonstrate the effectiveness of Tunsr.

Figures

Figures reproduced from arXiv: 2507.03697 by the authors.

Figure 1
Figure 1. Illustration of neural and symbolic methods for KG reasoning. Neural methods learn entity and relation embeddings [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of four reasoning scenarios on KGs: trans [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. An overview of the TUNSR. It utilizes multiple logic blocks to find the answer, where the reasoning graph is constructed and iteratively expanded. Meanwhile, a forward logic message-passing mechanism is proposed to update embeddings and attentions for unified propositional and FOL reasoning. (a) An example of reasoning graph in SKGs. (b) An example of reasoning graph in TKGs [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Examples of the reasoning graph with three iterations. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: The impacts of propositional and FOL reasoning on transductive, interpolation, and extrapolation scenarios. It is [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: The impacts of reasoning iterations which correspond to the length of the reasoning rules. It is evident that choosing [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: The impacts of sampling in the reasoning process. Performance and GPU space usage with batch size 64. Large [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

88 extracted references · 78 canonical work pages

  1. [1]

    Tiddi and S

    I. Tiddi and S. Schlobach. Knowledge graphs as tools for explain- able machine learning: A survey. Artificial Intelligence, 302:103627, 2022

    2022

  2. [2]

    Li and M

    M. Li and M. Moens. Dynamic key-value memory enhanced multi-step graph reasoning for knowledge-based visual question answering. In Thirty-Sixth AAAI Conference on Artificial Intelligence, pp. 10983–10992. AAAI Press, 2022

    2022

  3. [3]

    Mavromatis et al

    C. Mavromatis et al. Tempoqr: Temporal question reasoning over knowledge graphs. In Thirty-Sixth AAAI Conference on Artificial Intelligence, pp. 5825–5833. AAAI Press, 2022

    2022

  4. [4]

    Yang et al

    Y. Yang et al. Knowledge graph contrastive learning for recom- mendation. In The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 1434–

  5. [5]

    Zhu et al

    Y. Zhu et al. Recommending learning objects through attentive heterogeneous graph convolution and operation-aware neural network. IEEE Transactions on Knowledge and Data Engineering (TKDE), 35:4178–4189, 2023

    2023

  6. [6]

    Bastos et al

    A. Bastos et al. RECON: relation extraction using knowledge graph context in a graph neural network. In The Web Conference (WWW), pp. 1673–1685. ACM / IW3C2, 2021

    2021

  7. [7]

    Chen et al

    X. Chen et al. Knowprompt: Knowledge-aware prompt-tuning with synergistic optimization for relation extraction. In The Web Conference (WWW), pp. 2778–2788. ACM, 2022

    2022

  8. [8]

    B. D. Trisedya et al. GCP: graph encoder with content-planning for sentence generation from knowledge bases. IEEE Transactions on Pattern Analysis and Machine Intelligence (TP AMI) , 44(11):7521– 7533, 2022

    2022

Show all 88 references
  1. [9]

    Yu et al

    W. Yu et al. A survey of knowledge-enhanced text generation. ACM Comput. Surv., 54(11s):227:1–227:38, 2022

    2022

  2. [10]

    K. D. Bollacker et al. Freebase: A collaboratively created graph database for structuring human knowledge. In Proceedings of the International Conference on Management of Data (SIGMOD) , pp. 1247–1250, 2008

    2008

  3. [11]

    Vrandecic

    D. Vrandecic. Wikidata: A new platform for collaborative data collection. In Proceedings of the 21st World Wide Web Conference (WWW), pp. 1063–1064, 2012

    2012

  4. [12]

    Wang et al

    Q. Wang et al. Knowledge graph embedding: A survey of ap- proaches and applications. IEEE Transactions on Knowledge and Data Engineering (TKDE), 29(12):2724–2743, 2017

    2017

  5. [13]

    Rossi et al

    A. Rossi et al. Knowledge graph embedding for link prediction: A comparative analysis. ACM Transactions on Knowledge Discovery from Data (TKDD), 15(2):1–49, 2021

    2021

  6. [14]

    Pinker and J

    S. Pinker and J. Mehler. Connections and symbols. Mit Press, 1988

    1988

  7. [15]

    T. H. Trinh et al. Solving olympiad geometry without human demonstrations. Nature, 625(7995):476–482, 2024

    2024

  8. [16]

    Lin et al

    Q. Lin et al. Contrastive graph representations for logical formulas embedding. IEEE Transactions on Knowledge and Data Engineering , 35:3563–3574, 2023

    2023

  9. [17]

    Xu et al

    F. Xu et al. Symbol-llm: Towards foundational symbol-centric in- terface for large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , pp. 13091–13116, 2024

    2024

  10. [18]

    Lin et al

    Q. Lin et al. Fusing topology contexts and logical rules in language models for knowledge graph completion. Information Fusion , 90:253–264, 2023

    2023

  11. [19]

    Bordes et al

    A. Bordes et al. Translating embeddings for modeling multi- relational data. In Advances in Neural Information Processing Systems (NeurIPS), pp. 2787–2795, 2013

    2013

  12. [20]

    L. A. Gal ´arraga et al. AMIE: association rule mining under incomplete evidence in ontological knowledge bases. In 22nd International World Wide Web Conference (WWW), pp. 413–422, 2013

    2013

  13. [21]

    Yang et al

    F. Yang et al. Differentiable learning of logical rules for knowl- edge base reasoning. In Advances in Neural Information Processing Systems (NeurIPS), pp. 2319–2328, 2017

    2017

  14. [22]

    Shen et al

    Y. Shen et al. Modeling relation paths for knowledge graph completion. IEEE Transactions on Knowledge and Data Engineering , 33(11):3607–3617, 2020

    2020

  15. [23]

    Cheng et al

    K. Cheng et al. Rlogic: Recursive logical rule learning from knowledge graphs. In The 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) , pp. 179–189. ACM, 2022

    2022

  16. [24]

    Liu et al

    J. Liu et al. Latentlogic: Learning logic rules in latent space over knowledge graphs. In Findings of the EMNLP, pp. 4578–4586, 2023

    2023

  17. [25]

    Jiang et al

    C. Jiang et al. Path spuriousness-aware reinforcement learning for multi-hop knowledge graph reasoning. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL), pp. 3173–3184, 2023

    2023

  18. [26]

    Lin et al

    Q. Lin et al. Incorporating context graph with logical reasoning for inductive relation prediction. In The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pp. 893–903, 2022

    2022

  19. [27]

    Li et al

    J. Li et al. Teast: Temporal knowledge graph embedding via archimedean spiral timeline. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL) , pp. 15460–15474, 2023

    2023

  20. [28]

    Liu et al

    Y. Liu et al. Tlogic: Temporal logical rules for explainable link forecasting on temporal knowledge graphs. In Thirty-Sixth AAAI Conference on Artificial Intelligence , pp. 4120–4127. AAAI Press, 2022

    2022

  21. [29]

    Li et al

    N. Li et al. Tr-rules: Rule-based model for link forecasting on temporal knowledge graph considering temporal redundancy. In Findings of the Association for Computational Linguistics (EMNLP) , pp. 7885–7894, 2023

    2023

  22. [30]

    Lin et al

    Q. Lin et al. TECHS: temporal logical graph networks for ex- plainable extrapolation reasoning. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL) , pp. 1281–1293, 2023

    2023

  23. [31]

    Cambria et al

    E. Cambria et al. SenticNet 7: A commonsense-based neurosym- bolic AI framework for explainable sentiment analysis. In LREC, pp. 3829–3839, 2022. 14

    2022

  24. [32]

    Sadeghian et al

    A. Sadeghian et al. DRUM: end-to-end differentiable rule mining on knowledge graphs. In Advances in Neural Information Processing Systems (NeurIPS), pp. 15321–15331, 2019

    2019

  25. [33]

    Qu et al

    M. Qu et al. Rnnlogic: Learning logic rules for reasoning on knowledge graphs. In 9th International Conference on Learning Representations (ICLR), 2021

    2021

  26. [34]

    Zhang et al

    Y. Zhang et al. GMH: A general multi-hop reasoning model for KG completion. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pp. 3437–3446, 2021

    2021

  27. [35]

    Zhang et al

    J. Zhang et al. Subgraph retrieval enhanced model for multi- hop knowledge base question answering. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL), pp. 5773–5784, 2022

    2022

  28. [36]

    Lan et al

    Y. Lan et al. Complex knowledge base question answering: A survey. IEEE Trans. Knowl. Data Eng., 35(11):11196–11215, 2023

    2023

  29. [37]

    Dong et al

    H. Dong et al. Temporal inductive path neural network for temporal knowledge graph reasoning. Artificial Intelligence , pp. 104085, 2024

    2024

  30. [38]

    Chung et al

    J. Chung et al. Empirical evaluation of gated recurrent neural networks on sequence modeling. CoRR, abs/1412.3555, 2014

    2014 arXiv

  31. [39]

    Abiteboul et al

    S. Abiteboul et al. Foundations of databases , volume 8. Addison- Wesley Reading, 1995

    1995

  32. [40]

    Gebser et al

    M. Gebser et al. Potassco: The potsdam answer set solving collection. Ai Communications, 24(2):107–124, 2011

    2011

  33. [41]

    Alviano et al

    M. Alviano et al. Wasp: A native asp solver based on constraint learning. In Logic Programming and Nonmonotonic Reasoning: 12th International Conference, LPNMR 2013, Corunna, Spain, September 15-19, 2013. Proceedings 12, pp. 54–66. Springer, 2013

    2013

  34. [42]

    Rautenberg

    W. Rautenberg. A Concise Introduction to Mathematical Logic . Springer, 2006

    2006

  35. [43]

    Ciravegna et al

    G. Ciravegna et al. Logic explained networks. Artificial Intelligence, 314:103822, 2023

    2023

  36. [44]

    Ren and J

    H. Ren and J. Leskovec. Beta embeddings for multi-hop logical reasoning in knowledge graphs. In Advances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  37. [45]

    P . B. Andrews. An Introduction to Mathematical Logic and Type Theory: To Truth Through Proof , volume 27. Springer Science & Business Media, 2013

    2013

  38. [46]

    Sun et al

    J. Sun et al. A survey of reasoning with foundation models. arXiv preprint arXiv:2312.11562, 2023

    2023 arXiv

  39. [47]

    Zhang et al

    W. Zhang et al. Knowledge graph reasoning with logics and embeddings: Survey and perspective. CoRR, abs/2202.07412, 2022

    2022 arXiv

  40. [48]

    D. Poole. Probabilistic horn abduction and bayesian networks. Artificial Intelligence, 64(1):81–129, 1993

    1993

  41. [49]

    Xu et al

    D. Xu et al. Inductive representation learning on temporal graphs. In 8th International Conference on Learning Representations (ICLR) , 2020

    2020

  42. [50]

    Gal ´arraga et al

    L. Gal ´arraga et al. Fast rule mining in ontological knowledge bases with AMIE+. The VLDB Journal, 24(6):707–730, 2015

    2015

  43. [51]

    Zhang et al

    W. Zhang et al. Iteratively learning embeddings and rules for knowledge graph reasoning. In The World Wide Web Conference (WWW), pp. 2366–2377, 2019

    2019

  44. [52]

    Lacroix et al

    T. Lacroix et al. Canonical tensor decomposition for knowledge base completion. In Proceedings of the 35th International Conference on Machine Learning (ICML) , volume 80, pp. 2869–2878. PMLR, 2018

    2018

  45. [53]

    Yang et al

    B. Yang et al. Embedding entities and relations for learning and inference in knowledge bases. In International Conference on Learning Representations (ICLR), 2015

    2015

  46. [54]

    Xiong et al

    B. Xiong et al. Ultrahyperbolic knowledge graph embeddings. In The 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pp. 2130–2139. ACM, 2022

    2022

  47. [55]

    Wang et al

    J. Wang et al. Duality-induced regularizer for semantic matching knowledge graph embeddings. IEEE Transactions on Pattern Anal- ysis and Machine Intelligence (TP AMI), 45(2):1652–1667, 2023

    2023

  48. [56]

    Zhang et al

    Y. Zhang et al. Bilinear scoring function search for knowledge graph learning. IEEE Transactions on Pattern Analysis and Machine Intelligence (TP AMI), 45(2):1458–1473, 2023

    2023

  49. [57]

    Li et al

    R. Li et al. How does knowledge graph embedding extrapolate to unseen data: A semantic evidence view. In Thirty-Sixth AAAI Conference on Artificial Intelligence (AAAI) , pp. 5781–5791. AAAI Press, 2022

    2022

  50. [58]

    Zhang and Q

    Y. Zhang and Q. Yao. Knowledge graph reasoning with relational digraph. In The ACM Web Conference, pp. 912–924. ACM, 2022

    2022

  51. [59]

    Ge et al

    X. Ge et al. Compounding geometric operations for knowledge graph completion. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL), pp. 6947–6965, 2023

    2023

  52. [60]

    Wei et al

    W. Wei et al. Enhancing heterogeneous knowledge graph com- pletion with a novel gat-based approach. ACM Transactions on Knowledge Discovery from Data, 2024

    2024

  53. [61]

    Shi et al

    F. Shi et al. Tgformer: A graph transformer framework for knowledge graph embedding. IEEE Transactions on Knowledge and Data Engineering, 2025

    2025

  54. [62]

    L. A. Gal ´arraga et al. Amie: association rule mining under incom- plete evidence in ontological knowledge bases. In Proceedings of the 22nd international conference on World Wide Web , pp. 413–422, 2013

    2013

  55. [63]

    Meilicke et al

    C. Meilicke et al. Anytime bottom-up rule learning for knowledge graph completion. In IJCAI, pp. 3137–3143, 2019

    2019

  56. [64]

    Ott et al

    S. Ott et al. SAFRAN: an interpretable, rule-based link prediction method outperforming embedding models. In 3rd Conference on Automated Knowledge Base Construction (AKBC), 2021

    2021

  57. [65]

    Nandi et al

    A. Nandi et al. Simple augmentations of logical rules for neuro- symbolic knowledge graph completion. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL), pp. 256–269, 2023

    2023

  58. [66]

    Guo et al

    J. Guo et al. A unified joint approach with topological context learning and rule augmentation for knowledge graph completion. In Findings of the Association for Computational Linguistics , pp. 13686–13696, 2024

    2024

  59. [67]

    Teru et al

    K. Teru et al. Inductive relation prediction by subgraph reasoning. In International Conference on Machine Learning , pp. 9448–9457, 2020

    2020

  60. [68]

    Sun et al

    K. Sun et al. Incorporating multi-level sampling with adaptive aggregation for inductive knowledge graph completion. ACM Transactions on Knowledge Discovery from Data, 2024

    2024

  61. [69]

    Meilicke et al

    C. Meilicke et al. Fine-grained evaluation of rule-and embedding- based systems for knowledge graph completion. In 17th Interna- tional Semantic Web Conference, pp. 3–20, 2018

    2018

  62. [70]

    Mai et al

    S. Mai et al. Communicative message passing for inductive relation reasoning. In Thirty-Fifth AAAI Conference on Artificial Intelligence, pp. 4294–4302, 2021

    2021

  63. [71]

    Chen et al

    J. Chen et al. Topology-aware correlations between relations for inductive link prediction in knowledge graphs. In Thirty-Fifth AAAI Conference on Artificial Intelligence, pp. 6271–6278, 2021

    2021

  64. [72]

    Pan et al

    Y. Pan et al. A symbolic rule integration framework with logic transformer for inductive relation prediction. In Proceedings of the ACM Web Conference, pp. 2181–2192, 2024

    2024

  65. [73]

    Leblay and M

    J. Leblay and M. W. Chekol. Deriving validity time in knowledge graph. In Companion of the The Web Conference (WWW) , pp. 1771–

  66. [74]

    Goel et al

    R. Goel et al. Diachronic embedding for temporal knowledge graph completion. In The Thirty-Fourth AAAI Conference on Ar- tificial Intelligence, pp. 3988–3995, 2020

    2020

  67. [75]

    Garc ´ıa-Dur´an et al

    A. Garc ´ıa-Dur´an et al. Learning sequence encoders for temporal knowledge graph completion. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pp. 4816–4821, 2018

    2018

  68. [76]

    Sadeghian et al

    A. Sadeghian et al. Chronor: Rotation based temporal knowledge graph embedding. In Thirty-Fifth AAAI Conference on Artificial Intelligence, pp. 6471–6479, 2021

    2021

  69. [77]

    Lacroix et al

    T. Lacroix et al. Tensor decompositions for temporal knowledge base completion. In 8th International Conference on Learning Repre- sentations (ICLR), 2020

    2020

  70. [78]

    Xu et al

    C. Xu et al. Temporal knowledge graph completion using a linear temporal regularizer and multivector embeddings. In Proceedings of the 2021 Conference of the North American Chapter of the Asso- ciation for Computational Linguistics: Human Language Technologies (NAACL-HLT), pp....

    2021

  71. [79]

    Messner et al

    J. Messner et al. Temporal knowledge graph completion using box embeddings. In Thirty-Sixth AAAI Conference on Artificial Intelligence, pp. 7779–7787, 2022

    2022

  72. [80]

    Chen et al

    K. Chen et al. Rotateqvs: Representing temporal information as rotations in quaternion vector space for temporal knowledge graph completion. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL), pp. 5843–5857, 2022

    2022

  73. [81]

    Trouillon et al

    T. Trouillon et al. Complex embeddings for simple link prediction. In International Conference on Machine Learning (ICML) , volume 48, pp. 2071–2080, 2016

    2016

  74. [82]

    Jin et al

    W. Jin et al. Recurrent event network: Autoregressive structure inferenceover temporal knowledge graphs. In Proceedings of the 15 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 6669–6683, 2020

    2020

  75. [83]

    Zhu et al

    C. Zhu et al. Learning from history: Modeling temporal knowl- edge graphs with sequential copy-generation networks. In Thirty- Fifth AAAI Conference on Artificial Intelligence , pp. 4732–4740, 2021

    2021

  76. [84]

    Han et al

    Z. Han et al. Explainable subgraph reasoning for forecasting on temporal knowledge graphs. In 9th International Conference on Learning Representations (ICLR), 2021

    2021

  77. [85]

    Sun et al

    H. Sun et al. Timetraveler: Reinforcement learning for temporal knowledge graph forecasting. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pp. 8306–8319, 2021

    2021

  78. [86]

    Li et al

    N. Li et al. Infer: A neural-symbolic model for extrapolation reasoning on temporal knowledge graph. In The Thirteenth In- ternational Conference on Learning Representations (ICLR), 2025

    2025

  79. [87]

    Cambria et al

    E. Cambria et al. Seven pillars for the future of artificial intelli- gence. IEEE Intelligent Systems, 38(6):62–69, 2023

    2023

  80. [88]

    Xu et al

    F. Xu et al. Are large language models really good logical reason- ers? a comprehensive evaluation and beyond. IEEE Transactions on Knowledge and Data Engineering, 2025

    2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.